adversarial-feedback-chatbot

EMNLP 2020 findings paper "Learning Improvised Chatbots from Adversarial Modifications of Natural Language Feedback"

Data transformation with style-transformer

Based on style-transformer, we implement style-bart to incorporate bart model in transforming feedback conversation data into human bot conversation data.

Bart Model

The bart model is pre-trained with chatbot training data, so it can produce the response from the given context.

Notebooks

Notebooks are used to get several baseline scores of regex and polyencoder

Processed data

New chatbot training data samples transformed by style-bart from feedback data.